Loss of heterozygosity in childhood de novo acute myelogenous leukemia

The mechanisms of leukemogenesis in childhood acute leukemia remain largely unknown. The diagnostic karyotype is one of the most important determinants of outcome in children and younger adults with acute myelogenous leukemia (AML).1 Up to 50% of patients with AML have specific rearrangements such as t(8;21), t(15;17), and inv(16) in their leukemic blasts that correlate with characteristic clinical and biologic characteristics.2Creation of fusion genes by these and other translocations has been postulated to be a primary event in leukemogenesis.3 Because phases of progression are not generally apparent in de novo AML and additional genetic changes are often not observed on karyotype analysis, it has been proposed that additional mutations may not be necessary for leukemogenesis.3 However, the products of these fusion genes (AML/ETO, PML/RARA, andCBFA/MYH11) appear insufficient to cause AML.4-10 Other genetic changes are required for leukemogenesis, especially since approximately 20% of samples from patients with pediatric AML have no detectable karyotypic abnormality1 (and unpublished data).

In solid tumors, loss of heterozygosity (LOH) leading to loss of function of tumor suppressor genes plays a critical role in neoplastic progression and may have prognostic importance.11,12However, the role of LOH in the pathogenesis of leukemias is not as well defined. Several groups have identified regions of high-frequency allelic loss in childhood acute lymphoblastic leukemia (ALL).13-15 Partly on the basis of such studies, thep16/INK4A gene at 9p21 has been determined to be a likely tumor suppressor gene involved in childhood ALL.16,17Efforts are under way to identify a tumor suppressor gene on chromosome 12p13, possibly the TEL/ETV6 orp27/KIP1/CDNK1B gene18 or a novel gene in the vicinity.19 

Current evidence suggests that tumor suppressor genes may also be important in the pathogenesis of AML. A low level or lack of p27 (Kip1) protein appears to be an important clinical marker of AML progression.20 Similarly, methylation of thep15/INK4b promoter on 9p21 is commonly observed in ALL and AML and carries a worse prognosis.21,22 Monosomy 7 or partial deletions of the long arm of chromosome 7 (7q−), as well as partial deletions of the long arm of chromosome 5 (5q−), are nonrandom abnormalities observed in primary and therapy-induced myelodysplastic syndrome (MDS) and AML. The search for tumor suppressor genes in these regions is an active area of research.23-25 Germ-line inactivating mutations in the neurofibromin (Nf1) tumor suppressor gene create a predisposition to the development of AML.26-28 Loss of function mutations in another gene—CBFA2/AML1—appear to be responsible for the rare autosomal dominant syndrome of familial thrombocytopenia with a propensity to develop AML.29 Monoallelic mutations in theCBFA2/AML1 gene have been proposed to act by means of haploinsufficient tumor suppression. Point mutations in theCBFA2/AML1 gene have been described in about 5% of sporadic myeloid leukemias.30 

A systematic genome-wide screening for possible tumor suppressor genes in childhood AML that uses LOH results has not been reported. To investigate the prevalence of LOH in childhood de novo AML, we conducted an allelotype analysis using diagnostic bone marrow or peripheral blood samples from 53 patients with the disease. We used flow cytometry to obtain highly purified control lymphocytes and leukemic blasts for DNA extraction. We applied whole-genome amplification,31 which allowed us to conduct multilocus LOH assays using very small quantities of DNA. Using assays based on polymerase chain reaction (PCR), we screened each chromosome by using 177 highly polymorphic microsatellite repeat markers. We then used these findings to determine whether the presence of LOH was associated with specific clinical and biologic characteristics.

Cryopreserved bone marrow or peripheral blood specimens from 56 children with de novo AML were obtained from the Children's Cancer Group (CCG) AML Reference Laboratory (Fred Hutchinson Cancer Research Center, Seattle, WA). Three samples were excluded (those from patients 12 and 42, who had MDS, and that from patient 32, who had a high percentage of nonleukemic cells), leaving a total of 53 samples for analysis. Study participants were treated according to the CCG 2891 protocol,32 except for 5 patients who were treated according to the CCG 2961 protocol.33 Informed consent and assent was obtained from parents and patients as appropriate. Patients were selected on the basis of whether cryopreserved diagnostic bone marrow or peripheral blood was available for study. Results of karyotype analysis were available for all but 2 patients. The blast population from each sample was sorted on the basis of both forward (cell size) and right-angle light scatter (granularity). Diagnostic samples contained a high proportion of leukemic blasts (> 95% in most cases), which minimized contamination by normal blasts. For most patients, neither remission specimens nor other sources of constitutional DNA were available.

As a source of normal control DNA, we used highly purified, flow-sorted B and T lymphocytes (CD3 or CD20⁺ cells) from the diagnostic specimens. To permit multilocus LOH analysis of low numbers of cells, DNA from the sorted AML leukemic blasts and lymphocytes was subjected to primer extension preamplification (PEP) using random 15-mers for whole-genome amplification.34 35 DNA from at least 4000 cells was used in each PEP reaction, and the results of 3 duplicate PEP reactions were pooled. For 7 patients, we compared the results with those obtained by using DNA isolated from epithelial cells from the buccal mucosa.

CCG institutional laboratories conducted cytogenetic analyses by using standard techniques with an aliquot of the same diagnostic sample used for this study. Reporting forms including processing and analysis information as well as 2 original karyotypes from each abnormal clone (or of normal cells in the case of normal results) were centrally reviewed by at least 2 members of the CCG Cytogenetics Committee. Case analyses were considered adequate if an abnormal clone was defined or if 20 or more metaphase cells were analyzed and found to have a normal karyotype. Individual CCG institutions directly provided the karyotype information for 10 patients. Karyotypes were designated according to the International System for Human Cytogenetic Nomenclature (1995).36 

For this study, we used primer pairs located on all nonacrocentric chromosomal arms that amplified primarily trinucleotide and tetranucleotide repeat polymorphisms. Such primers have advantages over dinucleotide repeats, with lower strand slippage and wider allele separation.37 We screened all patients with the 167 fluorescent-labeled primer pairs contained in Research Genetics set 9A (Research Genetics, Huntsville, AL). Fifteen of these primers (GATA30D09, GATA89G08, D11S2000, DYS389, GGAAT1B07, D2S1384, D21S2052, GATA129D11, D11S968, D2S1328, GATA163B10, D17S928, GATA128C02, D16S2624, and D13S793) did not perform reliably. The single marker on the Y chromosome that performed reliably, DXYS154, detects a specific allele on the Y chromosome as well as on chromosome Xq28. Samples were also screened with the following primer pairs: D2S1788, D4S1627, D5S816, D5S1053, D7S1802, D7S1804, D8S1179, D9S921, D10S2325, D10S1237, D11S2362, D13S325, D14S587, D15S659, D18S535, D19S586, D20S604, D21S1437, D22S685, and D22S417 (Research Genetics). We also used one marker intragenic to the neurofibromatosis 1 gene,38 one intragenic to p53 (17pVNTR), and 3 markers (56-85, 57-71, and 57-76) intragenic to the CBFA2/AML1 gene.29A total of 177 primer pairs with an average spacing of 22 megabases (Mb) and an average heterozygosity of 0.8 were used for this analysis.

Locus-specific PCRs were carried out in a Tetrad DNA engine thermocycler (model PTC-200; MJ Research, Watertown, MA) by using a Taq PCR Master Mix kit (Qiagen, Valencia, CA), with PCR conditions recommended by Research Genetics and optimized for our laboratory. Primers were labeled with one of 3 fluorescent dyes (6-carboxy-fluorescein, hexachlorofluorescein, or 6-carboxy-4,7,2′,7′-tetrachloro-fluorescein). After locus-specific PCR, up to 10 products were pooled and run with an internal size standard (Genescan-500 labeled with 6-carboxy-N,N,N′, N′-tetramethylrhodamine; Perkin Elmer Applied Biosystems, Foster City, CA) in each lane on an ABI Prism 377 DNA sequencing machine (Perkin Elmer Applied Biosystems). Data were analyzed by using ABI GeneScan software (Perkin Elmer Applied Biosystems) with visual quality control of the fluorescent histogram.39 For informative cases, a quantitative LOH (QLOH) score was determined as the ratio of the blast to lymphocyte allele peak height and expressed as a value between 0 and 1. Samples with a QLOH of 0.5 or lower were initially scored as positive for LOH, although patients whose karyotype indicated an allelic imbalance due to trisomy were not scored as having LOH (Table1). All LOH events were confirmed by using lymphocyte and blast DNA from the same patients without PEP. We used the genetic location database of the Wessex Human Genetics Institute, University of Southampton, United Kingdom (http://cedar.genetics.soton.ac.uk/public_html/),40 to assign markers to their most likely cytogenetic band. The Unified Database for Human Genome Mapping41 was used to determine the distance in Mb between markers.

Demographic and clinical data on the patients in the study were collected with quality control review done according to the standard procedures of the CCG. Standard statistical tests, including χ² tests for differences in proportions and Mann-Whitney rank-sum tests, were used to determine whether the samples chosen for study were representative of the CCG 2891 population with respect to age, sex, white blood cell (WBC) count at diagnosis, French-American-British (FAB) classification, cytogenetic subgroup, and induction outcomes and to test for associations between these clinical characteristics and LOH. Complete data for these comparisons were available for 47 of the 53 patients. Standard criteria were used to define clinical remission. Overall survival (OS) was measured from random assignment at diagnosis until death from any cause, with observations censored for patients last known to be alive. Distributions of OS were estimated by using the Kaplan-Meier method.42 Significance was estimated with use of 2-tailedP values. Analyses were based on clinical and biologic data available at the time of analysis.

No discrepancies were observed between DNA derived from buccal mucosa samples and that derived from lymphocytes among 586 total alleles from 7 patients (patients 30-36); 452 of these alleles were informative (data not shown). LOH in 2 of these samples (from patients 30 and 35) was confined to the leukemic blasts (Table 1). Furthermore, in 3 patients in whom trisomy was observed in the leukemic blasts on karyotype analysis (patients 30, 35, and 36), a comparable degree of allelic imbalance was detected in comparing leukemic DNA with that derived from either buccal mucosa or lymphocytes. In addition, we could detect nearly all chromosomal deletions for which we had reliable informative markers and that were present in most cells karyotyped. These findings indicated that flow-sorted lymphocytes were a valid source of control DNA for these studies.

DNA from flow-sorted leukemic blasts and matching flow-sorted lymphocytes from 53 patients were analyzed for LOH by using 177 highly polymorphic microsatellite repeat markers. Informative allelotypes were found in 6651 of 8904 sites (75%) examined. A total of 19 LOH events were identified in 17 patients, on chromosome arms 1q, 3q, 5q, 7q (n = 2), 9q (n = 4), 11p (n = 2), 12p (n = 3), 13q (n = 2), 16q, 19q, and Y (Table 1). In most cases, one allele was completely absent, indicating allelic loss in a clonal cell population and little contamination by normal cells. Ninety-eight percent of the chromosome arms tested (2076 of 2118) had at least one informative marker that yielded a mean fractional allelic loss (FAL; ie the ratio of chromosomal arms showing allelic loss to the number of informative arms in each patient) of 0.9%. There was no association between a specific AML FAB subtype and the presence of LOH in this group (Table 1). LOH was not found by using intragenic markers for p53,Nf-1, or CBFA2/AML1. True microsatellite instability is rare in de novo acute leukemias43-45 and was not detected in any of the samples analyzed with our markers. However, we did observe 3 instances in which one of the amplified alleles in the blast DNA was 4 base pairs (bp) larger than that obtained by using lymphocyte DNA (in patient 10 [D1S1609], patient 8 [D20S480], and patient 30 [D4S2639]) and one instance of a 4-bp deletion in one of the blast alleles (in patient 4 [D16S753]). These results were verified by using DNA from the same patients without PEP.

The LOH findings were compared with the results of cytogenetic analyses (Table 1). Nine of the 19 LOH events were not detected by standard cytogenetic analyses; these included the events on 1q, 3q, 5q, 13q (n = 2), 11p (n = 2), 12p, and 19q. We did not have an informative marker in the deletion intervals on 7q and 11q in patient 3 or reliable Y chromosome results for patient 53, who had a Y chromosome deletion. The only cytogenetic deletion in this group that was present in most cells and for which we had an informative marker in the deletion interval and failed to detect LOH was in patient 52. This patient had a Y chromosome deletion in 16 of 20 cells, but 2 alleles were detected by using marker DXYS154, which recognizes specific alleles on both Y and Xq.

Our results clearly showed the value of having a comparison karyotype in LOH analyses. Allelic imbalances detected by LOH studies do not always indicate the loss of an allele. Trisomy results in an allelic imbalance of 1:2, which should theoretically yield a QLOH score of 0.5. However, the presence of normal blast cells or slight variations in the amplification of DNA could affect the QLOH score obtained. In this study, which defined LOH as a QLOH of less than 0.5, the mean QLOH scores for the informative markers in patients with trisomies ranged from 0.5 to 0.6, whereas those for samples thought to have true LOH ranged from 0 to 0.4 (Table 1).

To determine whether regions of LOH were not detected on routine karyotype analysis because of their short length, we analyzed samples without cytogenetically evident deletions by using additional flanking markers (data not shown). Three instances of LOH were detected only with a single informative primer pair (in patients 28, 47, and 41). The involved length of chromosome 3q in patient 4 was at least 19 Mb (D3S2427-D3S2418). In patient 43, the involved region on chromosome 5q was at least 49 Mb (D5S2501-D5S1456). The involved region on chromosome 11 was at least 23 Mb in patient 1 (D11S1984-ATA34E08) and 36 Mb in patient 2 (D11S1984-D11S1392). Patient 25 had detectable LOH for all informative markers tested on the long arm of the acrocentrometric chromosome 13, spanning at least 67 Mb (D13S1493-D13S895), whereas in patient 21, LOH was present for at least 19 Mb on chromosome 19q (D19S431-D19S246). The finding of LOH at these loci by using multiple adjacent markers also indicates that the results were not simply an artifact due to unequal amplification of alleles during whole-genome amplification.

The age, sex, WBC count at diagnosis, and remission induction rate in the 47 patients selected for the LOH study who were treated according to the CCG 2891 protocol were compared with the entire study population (Table 2). No significant differences between the 2 groups in sex or OS were observed. The group selected for our analysis did have a higher median WBC count at diagnosis (40 × 10⁹/L versus 20 × 10⁹/L;P = .010 on Mann-Whitney U testing). Such patients are likely overrepresented in our reference bank. Infant leukemias (diagnosed in patients under 2 years of age) were not represented in our group of patients. Samples from this young group of patients are underrepresented in our reference bank because of problems collecting optimal sample volumes for banking. The analysis revealed that in the selection of patients for this study, there was a bias toward those with a lower rate of remission induction (P = .0025 on χ² testing). When we compared these characteristics in patients with and without LOH (Table 2), we found a significant association between LOH and a WBC count of 20 × 10⁹/L or higher (P = .035 on χ² testing). The median WBC count at diagnosis was also higher in patients with LOH (62 × 10⁹/L versus 29 × 10⁹/L) although not significantly (P = .06 on Mann-Whitney U testing). No significant differences in age at diagnosis, sex, or outcome in groups with or without LOH were apparent for OS (8-year follow-up) from enrollment in the study (51% ± 15% versus 34% ± 8%). This remained true when results were stratified according to WBC count (Table 2).

To locate possible tumor suppressor genes in patients with childhood de novo AML, we conducted a genome-wide evaluation of allelic loss in 53 diagnostic bone marrow or peripheral blood samples by using 177 polymorphic markers. In the absence of remission samples or other sources of comparison constitutional DNA, we did these studies by using amplified DNA of normal lymphocytes extracted from diagnostic samples from patients with AML. The studies identified a total of 19 sites of LOH in 17 of 56 samples (32%). Allelic loss was identified on chromosome arms 1q, 3q, 5q, 7q (n = 2), 9q (n = 4), 11p (n = 2), 12p (n = 3), 13q (n = 2), 16q, 19q, and Y. Patient 25 and patient 3 each had 2 distinct sites of LOH (Table 1). Nine instances of LOH were not detected by classic cytogenetic analysis. Three of these cases were detected by a single primer; in the other 6, LOH was detected when multiple adjacent markers spanning at least 19 to 67 Mb were used. The lack of detection of these large regions of allelic loss by cytogenetic analysis represents either a skewing toward normal cells expanded for karyotype analysis or uniparental isodisomy, ie, a single set of duplicated alleles within these chromosomal regions. This could result from a deletion or duplication event or a trisomy with mitotic recombination and subsequent loss of one chromosome. LOH analyses are uniquely suited for identifying such events undetectable by other techniques, such as high-resolution karyotyping, spectral karyotyping, or comparative genome hybridization.

This study clearly showed the value of having a comparison karyotype in LOH analyses. With a cut-off value of 0.5 for QLOH, several instances of trisomy would have been inappropriately classified as LOH. Other mechanisms, such as amplification and chromosomal reduplication, can also result in an allelic imbalance. QLOH values between 0.2 and 0.5 may be due to such events or to partial contamination of tumor cells with normal cells. In most instances of LOH detected in this study, there was complete loss of one allele. Many studies of LOH do not reveal the QLOH value or karyotype and thus fail to rigorously differentiate LOH from other possible causes of allelic imbalance. Also, in searching for possible sites of tumor suppressor genes, a karyotype can be useful for detecting regions of chromosomal loss missed by allelotyping because of a lack of informative alleles in the deletion interval or because deletions are present in a minority of cells.

Our study did not find LOH involving 2 genes, Nf1 andAML1/CBFA2, whose functional loss predisposes to the development of AML. We also did not find LOH of p53, a gene that is frequently abnormal in solid tumors. In the development of esophageal adenocarcinoma, disruption of p53 typically precedes the development of genetic instability including aneuploidy.11 Similarly, in therapy-related MDS and AML in adults, LOH of p53 is associated with a more complex karyotype.46 The absence of p53 LOH in pediatric de novo AML correlates with the relative preservation of genomic stability in this disease.

A much higher rate of LOH has been found in solid tumors and in studies of other leukemias. Esophageal adenocarcinomas have an FAL (ie, the ratio of chromosomal arms showing allelic loss to the number of informative arms in each patient) of 26% to 34% in early and advanced disease.34 Studies of childhood ALL found FAL rates of 3% to 12.3%.14,15 One study of adult AML using only 22 microsatellite markers found LOH in 25 of 32 samples.44 In contrast, our study found a mean FAL of 0.9% (median, 0) in pediatric de novo AML. These results suggest that tumor suppressor genes play a lesser role in the pathogenesis of pediatric AML than in that of solid tumors and other leukemias. Other genetic aberrations, including fusion genes that impair differentiation and oncogenes that drive proliferation, may primarily underlie the development of pediatric AML. Alternatively, tumor suppressor genes in pediatric AML may be inactivated by other means, such as methylation or point mutations.

It is possible that the relatively few sites of LOH detected in this study represented random background events. However, the limited LOH detected in the absence of genetic instability could indicate rare sites of biologic importance in the pathogenesis of pediatric leukemia. For example, it was noted previously that chromosome arm 12p is frequently aberrant in childhood ALL,14,15,47 as is 13q in childhood ALL and adult chronic lymphocytic leukemia.14,48-50 Furthermore, several of the loci detected in this study were previously implicated in the pathogenesis of AML, including deletion of 9q,51 monosomy 7, partial deletions of 7q23,52-55 and 5q, and abnormalities of 3q.1 25 Our finding of LOH on 5q and 3q in 2 patients without such abnormalities on cytogenetic analysis indicates that LOH analysis may be a more sensitive technique for identifying such patients with a poor prognosis. The relative absence of individual high-frequency sites of loss in this study of AML may be a reflection of the heterogeneity of the disease.

The association we observed between LOH and an increased WBC count must be verified. However, such an association might occur if the presence of LOH defined a subgroup of AML in which the proliferative potential was higher. Although a WBC count of 20 × 10⁹/L or higher has been shown to be a poor prognostic factor in pediatric AML,56 we did not find a worse prognosis in patients with such blood counts in whom LOH was detected. This could have been a reflection of inadvertent bias in our study toward patients with a rate of remission induction lower than that in the general treatment group. Analogous to the differing prognostic categories associated with different cytogenetic abnormalities, it also seems likely that only certain sites of LOH would have prognostic importance. This issue may be resolved by a larger study focusing on specific sites of LOH.

This study represents the first reported genome-wide scanning for allelic loss in patients with childhood de novo AML and shows that LOH occurs with a relatively low frequency in the disorder. We also found that most mature lymphocytes are unaffected by deletions evident in the leukemic blasts and that comprehensive allelotyping can be achieved by using flow-sorted cells and whole-genome amplification of DNA from small, highly purified samples. Additional studies are needed to determine whether specific sites of LOH in fact harbor tumor suppressor genes and whether they correlate with specific biologic and clinical characteristics.